Migrating JGroups from 3.6.20 to 5.4.8. Question about RouterStubManager CPU consumption

[len-ro]

Hello, I am finally migrating from Jgroups 3.6.20 to 5.4.8 and I have a question about RouterStubManager. We are using GossipRouters on all major nodes and experienced a high CPU consumption in RouterStubManager when a node was down. Tracked this to a for loop in RouterStubManager. In the while loop some sleep would be needed. I saw that RouterStubManager is re-written in 5.4.8, can a similar issue occur, any feedback on this topic from your production envs?
I know this is a very specific question, but thank you anyway.

Regards

[belaban]

The sleep you're referring to is reconnect_interval (default: 0, no sleep). You could set it to 300, for example, and then each loop would sleep for 300ms before starting at the top again.
Also, if you have configured a sock_conn_timeout, this will wait for N ms until a socket connect times out.

[len-ro]

Thank you very much for your reply.

[belaban]

NP, let me know if this works

[len-ro]

By the way I was looking for migration documents for 3 to 4 and 4 to 5 but could not find anything structured. Is there anything? For instance I noticed that the viewAccepted is no longer received by the first node in the cluster.

Regards,

[belaban]

Not a real migration document, but read the 5.x manual and you should see the difference to 3.x. Unless you wrote custom protocols, migration should be simple.
Ask questions if you run into issues. Best accompanied by reproducers... :-)